12-05-2017, 05:27 PM | #1 |
Join Date: Dec 2017
|
Ogres too timid in Campaign
I'm playing scenario 8 in the Campaign. The train has to wait 8 turns before it leaves, and the Ogres have to try to kill it. However, in the face of 2 groups of Superheavies and several howitzers pre-placed in the scenario, the Ogres have backed off to a safe space 9 hexes from the howitzers. They appear to be willing to wait there until the train leaves and I win.
This is not a playable scenario against the AI. Anybody have any different experience? |
12-05-2017, 06:10 PM | #2 | |
Ogre Line Editor
Join Date: Sep 2006
Location: Plainfield, IL
|
Re: Ogres too timid in Campaign
Quote:
__________________
GranitePenguin Ogre Line Editor |
|
12-05-2017, 07:52 PM | #3 |
Join Date: Dec 2017
|
Re: Ogres too timid in Campaign
And in Scenario 9, after destroying the broadcast centers, I had 1 GEV remaining in the west, who went all the way up to the northwest corner, chased by all 3 enemy Ogres. This allowed my Ogre to march unopposed to within missile range of the CP and blow it to smithereens.
The AI needs to be programmed with some sense of the objectives they are to protect. |
12-06-2017, 01:36 AM | #4 |
Join Date: Jun 2012
|
Re: Ogres too timid in Campaign
It seems to play as if it has no idea what my victory conditions are, and that it can win by preventing me from achieving them.
If I win by destroying buildings, protect the buildings. If I win by destroying Ogres, keep the Ogres out of engagement. |
12-06-2017, 04:24 AM | #5 |
Moderator
Join Date: Jun 2017
Location: Bristol, UK
|
Re: Ogres too timid in Campaign
Thanks for the feedback on this folks - I've passed it on and in the next round of updates, we'll take a look at how this can be improved.
Thank you |
12-07-2017, 07:04 AM | #6 |
Join Date: Oct 2017
|
How to improve the a.i., part one
(This is for the long term. Right now fixing bugs ad adding elements that are in the board game is a much higher priority.
In have a lot of experience programming A.I.s for Chess. I think you could learn a few things from the way the chess programs do it. First, I am assuming that the A.I. looks at the result depending on what it does, and picks the move that it thinks best. In Chess, the A.I. typically assings 1 poit to a pawn (you gain a point if you take the opponent's pawn, lose a point if he takes yours) 3 points to a bishop, 9 to a Queen, and 9999 to a checkmated King (its OK to lose everything if by doing so you achieve checkmate). Controlling the center of the board gives you a fraction of a point, a queen that is trapped loses a fraction of a point, etc.) If the Ogre A.I doesn't follow this basis scheme, let me know and I will adjust my advice accordingly I assume that the game does something similar; killing a super heavy tank gives you more points than killing a light tank. So the first thing to do in the essential personnel scenario is to adjust the points so that killing the train or the track is a very high priority. Likewise, in the lucky break scenario killing infantry should give more points than normal, and killing infantry that is close to the disabled Ogre should be worth even more points. |
12-07-2017, 07:13 AM | #7 |
Join Date: Oct 2017
|
How to improve the a.i., part two
The next step is to implement a lookahead strategy. Instead of the A.I. looking at the result depending on what it does and picking the move that it thinks best, it considers all of the possible responses (including weighing the die roll - a result that needs a 2 or better is more likely than a result that requires rolling a 6) then picking the best move assuming that theb opponent pick his best move. The real powe5 comes into play when you go one step further and consider all of your possible responses. a 1, 2, or 3 move lookahead can usually be done in seconds.
|
12-07-2017, 07:17 AM | #8 |
Join Date: Oct 2017
|
How to improve the a.i., part three
Once you have the lookahead working, the next step is to tweak the point assigned for various factors. On good way to do this is to play thousands of A.I. vs A.I games with the A.I.s using slightly different point values. Losers get deleted, winners go on to future rounds.
|
12-07-2017, 08:25 AM | #9 |
Join Date: Jun 2008
Location: Cheltenham, PA
|
Re: Ogres too timid in Campaign
Based on my playthroughs, the #1 thing that I would suggest looking at is whether the AI takes into account both its own objectives AND its opponents, or just its own (and how they're weighted). Because even if its own objectives are heavily weighted, if preventing its opponents objectives are not equally weighted you get what we're seeing. In each scenario there's effectively 2 sets of objectives running in parallel - you need to reach your goal, while at the same time preventing your opponent from reaching their goal. The AI seems to be focused entirely on the former, and doesn't really pay attention to the latter...
Also, in skirmish games where the AI plays the defense, there doesn't seem to be any attempt to conserve and/or concentrate forces for maximum effect. This makes it easy to pick off individual units without much risk to the Ogre, eventually rendering the defense toothless...
__________________
Joshua Megerman, SJGames MIB #5273 - Ogre AI Testing Division |
12-07-2017, 09:51 AM | #10 |
Join Date: Dec 2017
|
Re: Ogres too timid in Campaign
Having the ability to modify objectives based on events would be helpful, too. In the scenario The Train, the AI rightly goes after the train tracks. But it continues to target train tracks after the train has escaped. At that point, the only thing that counts is killing enemy units for points.
I also watched a close game go heavily in my favor as 6 AI GEVs overran my city-placed Infantry(3), one at a time. I killed them all with no losses. If they had come at me all at once, I would have killed 1 or 2 and then been wiped out. The odds were terrible for the lone GEVs, yet they came at me singly and wiped out most of what the AI had left. |
Thread Tools | |
Display Modes | |
|
|